Miscommunication can quickly lead to accidents resulting in death or injury. The E-µ armband is a wearable device which serves as a supplemental interface between two people, with the purpose of reducing communication errors. As a test case, we assist in navigating large vehicles through difficult or heavily populated terrain. The device interprets a gesture by analyzing muscle activity and motion of the wearer’s arm on an embedded processor. Then, a voice command associated with the gesture is sent wirelessly over a Bluetooth connection to a speaker in the driver’s car. E-µ Armband Christopher Allum, Jeffrey Maloney, Shehzeen Hussain Faculty Advisor: Prof. Baird Soules Department of Electrical and Computer Engineering ECE 415/ECE 416 – SENIOR DESIGN PROJECT 2015 College of Engineering - University of Massachusetts Amherst SDP15 Abstract Block Diagram System Overview Acknowledgements We could not have done this project without the support of UMass faculty, friends and alumni. We would like to give special thanks to: -Engineer Michael Mckinley from U.S. Bionics -Professor Umberger from the Kinesiology Dept. -Our advisors, Professors Krishna and Aksamija -Our awesome friends Omid Meh and Shamit Som A wearable device to detect hand signals and relay information to a vehicle operator via recognizable pre- recorded voice commands. The E-µ project consists of two major interfaces: ARM BAND ● Inertial Measurement Unit (IMU): measures acceleration and angular velocity and sends data to the processor using the I 2 C protocol. ● Electromyography Sensor Circuitry: muscle signals are collected through reusable surface electrodes and then conditioned with analog filters and amplifiers before being fed to processor. ● Electromyography Signal Processing System: Fast Fourier Transform is performed on signals using the processor and features describing the signal are extracted from time and frequency domain. ● Bluetooth module: interfaces between microcontroller and car speaker. ● Two 3.7V Lithium Ion Battery Packs: rechargeable batteries. OUTPUT An audio signal delivered by a speaker inside the car upon command from arm band. ● Car power adapter: powers the Raspberry Pi ● Raspberry Pi: processes Bluetooth message and outputs stored voice commands to speaker Specifications SpecificationValue Weight1.5 kg Height2 cm Length40 cm Battery Life> 8 hours Range> 10 m Hardware Cost$240 Results ForwardReverseLeftRightStopSlow User 1100% 80%100% 80% User 280%60%80%100%80% User 3100% 40%60%100% Overall93%87%67%87%93%87% The system was tested by three users. Each user recorded five separate sets of each gesture and tested the success. Success was defined as the system outputting the correct command in under two seconds. Arm Band Output Device MCU EMG DSP IMU DSP Controller Pattern Recognition Power Supply 3.7V Battery 3.7V Battery +3.3V Reg -3.0V Reg EMG Signal Conditioning Amplifier Feed- back LPF Surface Electrodes IMU Sensor BT Module Legend +3.3V -3.0V Analog Signal Digital Signal Bluetooth Signal Software Block Hardware Block Raspberry Pi Speaker 5V Car Charger Mode Select Pushbutton Piezo Beeper

EMG Digital Signal Processing Recorded Data and PCA Fig: MATLAB Unprocessed Signal FFTFig: Teensy Processed Signal FFT Frequency Magnitude EMG signals provide distinctive features in both time and frequency domain. EMG signals are composed of action potentials fired by muscle fibers when performing gestures. The features are extracted and compared for classification of samples and building of a gesture library. ● Mean Amplitude (Time & Frequency Domain): Captures signal strength. ● Root Mean Square of Amplitude (Time & Frequency Domain): Power of the signal calculated by squaring each data point, summing the squares, dividing the sum by the number of observations, and taking the square root of result. ● Number of Zero Crossings (Time Domain): Counting the number of times the amplitude of the signal crosses the zero line. A more active muscle will generate more action potentials causing more zero crossings in the signal. ● Mod Frequency (Frequency Domain): Fast Fourier Transformation is used to break the EMG signal into its frequency components. Mod frequencies detect highest peaks. EMG DSP Time Domain Feature Extraction Pattern Recognition Subsystem FFT to Frequency Domain Fs = 2kHz, N =256 Frequency Domain Feature Extraction EMG Conditioning Circuitry Digital Noise Filter Data Signal Hardware Block Software Block Legend Forward Backward Left Right Stop Slow Pattern Recognition Overview: The pattern recognition system uses the K-Nearest Neighbors Algorithm (KNN) to map the wearer’s motions to gestures. It takes in IMU data and EMG data and outputs a score for each gesture. Causes of variability in gesture data include: - Unique muscle characteristics - Differing arm length between users - Inconsistent sensor placement To overcome this variability, we record new gestures upon each use. This allows for a more robust system. After completing a recording, all of the data is sent through a post-processing algorithm which removes outliers from the sample space. Pattern Recognition Stored Patterns Calculate Distances EMG/IMU Data Get N Nearest Neighbors Scoring Algorithm Moving Window Pattern Scores Sort Distances Data Signal Software Block Legend When in run mode, the user performs gestures that are recorded into the twelve dimensional signal space. While in run mode, the real time measurements are mapped to the same space and then compared against the recorded data. Pictured above is a three dimensional representation of a typical recorded data set and representation of greatest variance of each gesture compared to that of all gestures performed. EMG Sensor Circuitry The EMG circuitry conditions electrical signals from the surface of the skin for processing. Electrodes are placed on upper forearm near the elbow. These signals must be amplified independently from the other low frequency noise present at the surface of the skin. EMG Amplifier InAmp Anti-Aliasing LPF Feedback LPF MCU A/D Legend Analog Signal Hardware Block GUI/Processing Visualization Cost Recorded Patterns Projected into PCA Space PartDevelopment CostProduction Cost IMU Electronics PCB MCU Batteries BT Module Total Audio Output Feedback Vehicle operator receives audio commands inside the vehicle up to 10 meters in range from ground guide. Gestures are translated to commands sent wirelessly over Bluetooth link to car driver. BlueSMiRF Gold Bluetooth Module: ● Processor sends commands to receiver via Bluetooth Raspberry Pi connects to speakers inside car: ● Raspberry Pi powers from car charger and pairs wirelessly with E-µ. ● Python code plays sound files from SD card, based on the command